Exponential Smoothing of Realized Portfolio Weights

Abstract

In this paper we investigate exponential smoothing (ES) predictors for the weights of high-dimensional realized global minimum variance portfolios (GMVP) which only depend on the realized covariance matrix of risky financial assets. We compare direct ES predictions of realized GMVP proportions and indirect ES forecasts, where smoothing is applied to the realized covariance matrices and the GMVP composition is computed afterwards. We provide analytical results which show that either direct or indirect ES predictors of the GMVP proportions can be advantageous but neither of them dominates. For this reason we suggest a dynamic time series approach in order to combine them. We illustrate our findings in an empirical study for GMVPs based on 100 risky assets and report that the proposed ES forecast combination is suitable for GMVP prediction.